Exposing ML Workbench Prediction Results in Studio Without Manual Data Export

The simplest way to resolve your issue is to send the dataframe back to your data pool. Once there, you can use the data jobs to get a table in the data model so you can freely use your output from the MLWB within Studio.

To get it done, you can use the PyCelonis library! There’s a method specif to perform data push to the data pool :)

https://celonis.github.io/pycelonis/2.15.1/reference/pycelonis/

Hi Sagar,

You can follow the steps above if you want to load bigger data sets, and your use case does not require instant results (I.e. it’s okay that a data model load is done in between). PyCelonis is the way to go there.

For instant use case, you can populate augmented attributes using the PyCelonis/Celonis API directly, without the need for a data model reload. 
 

Best regards,

Jan-peter 

Thanks for the earlier suggestion about using PyCelonis to push MLWB prediction results back into a Data Pool. I did some deeper testing and found something that I'm hoping someone can clarify. I'm using ML Workbench (PyCelonis 2.15.1) with a Knowledge Model as the source. Reading data from the Knowledge Model, training/scoring a scikit-learn model, and generating Pandas prediction DataFrames all work successfully. However, when I try to access Data Integration through PyCelonis, celonis.data_integration.get_data_pools() returns an empty list ([]), and celonis.data_integration.get_data_pool("<actual_pool_id>") returns PyCelonisPermissionError: You don't have permission to perform 'GET'. I also confirmed that celonis.data_integration.create_data_pool() returns a permission error. The interesting part is that I can access the same Data Pool from the Celonis UI and the Data Pool metadata shows permissions such as VIEW_DATA_POOL, EDIT_DATA_POOL_RESTRICTED, and ADMIN for my user. However, when I check permissions from MLWB using celonis.team.get_permissions(), I see storage-manager permissions: [], and it appears that MLWB is running under a workload identity rather than directly under my interactive user account. My goal is to persist prediction results generated in MLWB back into Celonis so they can be consumed by a Knowledge Model and visualized in Studio (Knowledge Model → MLWB Prediction → Persist Results → Knowledge Model → Studio) without using manual CSV export/import. Has anyone successfully written a Pandas DataFrame from MLWB back into Data Integration using PyCelonis? Does the MLWB workload identity require separate permissions from the permissions assigned to my user account? Are specific Data Push API permissions required? Or is there a different Celonis-native pattern for persisting MLWB prediction results so they can be consumed by Knowledge Models and Studio? Any guidance would be greatly appreciated.

HI ​@Sagar_MTB ,

With regards to your permission error, you can :

1. provide access to your ML application in Data Integration : Data Integration → Data Pool Permissions → search for the name of your Workench you created in MLWB → grant read access. That’s similar to what you would do to provide a specific user/group permission to one data pool, OR
2. Since you have admin access, provide access to that application in Admin & Settings → Permissions

For the data push, I would try the create_table function explained in Data Upload & Export Advanced - PyCelonis.

I hope that helps!

Cheers,

Joao

Hi Joao, Yes now i am able to push the data from the MLWB back into the datapools now, after i have the data pool permissions. Thank you for the reply and everyone who replied regarding this topic. I really appreciate it!